The paper addresses a fundamental question in the study of certain nonlinear partial differential equations: if you know that solutions are smooth in a qualitative sense (meaning smoothness holds pointwise, without any measurements of how smooth), can you automatically conclude that they satisfy strong, uniform, quantitative estimates? The answer the authors establish is yes, under a natural set of conditions. Specifically, they work with a broad class of "fully nonlinear elliptic equations," which are equations where the relationship between a function and its second derivatives is complex and nonlinear, and which appear throughout geometry, physics, and optimization.
The core technical achievement is showing that qualitative smoothness and quantitative regularity share the same threshold for these equations. In practical terms, this means that if solutions in a natural family of such equations are twice differentiable everywhere in even a basic sense, then they automatically satisfy a much stronger, uniformly controlled smoothness condition known as C^{2,alpha} regularity. This is surprising because qualitative smoothness is a weak assumption that carries no numerical bounds, yet the conclusion is a sharp, measurable estimate. The authors also derive a partial regularity result: at any point where a solution happens to be twice differentiable, it is guaranteed to be fully regular there, while the remaining "bad" points where regularity might fail form a set of strictly lower dimension than the surrounding space.
The proof strategy brings together two key ideas. First, the authors use compactness arguments, meaning they work with the collection of all rescaled versions of the equation simultaneously and exploit the fact that this collection behaves well under limits. Second, they use a technique called improvement of flatness, which iteratively refines how closely a solution resembles a simple polynomial at smaller and smaller scales. Crucially, the argument never needs to quantify how smooth the solutions are to begin with, only that smoothness holds. This opens doors to problems where smoothness can be observed or inferred in principle, but where deriving explicit estimates has previously been out of reach.